ABUNDANCE RATIOS AND GALACTIC CHEMICAL EVOLUTION

Transcription

1 Annu. Rev. Astron. Astrophys : Copyright c 1997 by Annual Reviews Inc. All rights reserved ABUNDANCE RATIOS AND GALACTIC CHEMICAL EVOLUTION Andrew McWilliam Carnegie Observatories, Room 33, 813 Santa Barbara Street, Pasadena, California 91101; andy@marmite.ociw.edu KEY WORDS: abundances, chemical composition, the Galaxy, nucleosynthesis, stars ABSTRACT The metallicity of stars in the Galaxy ranges from [Fe/H] = 4 to +0.5 dex, and the solar iron abundance is ɛ(fe) = 7.51 ± 0.01 dex. The average values of [Fe/H] in the solar neighborhood, the halo, and Galactic bulge are 0.2, 1.6, and 0.2 dex respectively. Detailed abundance analysis reveals that the Galactic disk, halo, and bulge exhibit unique abundance patterns of O, Mg, Si, Ca, and Ti and neutron-capture elements. These signatures show that environment plays an important role in chemical evolution and that supernovae come in many flavors with a range of element yields. The 300-fold dispersion in heavy element abundances of the most metal-poor stars suggests incomplete mixing of ejecta from individual supernova, with vastly different yields, in clouds of 10 6 M. The composition of Orion association stars indicates that star-forming regions are significantly self-enriched on time scales of 80 million years. The rapid selfenrichment and inhomogeneous chemical evolution models are required to match observed abundance trends and the dispersion in the age-metallicity relation. INTRODUCTION Except for the lightest elements, the history of the chemical composition of the Galaxy is dominated by nucleosynthesis occurring in many generations of stars. Stars of low mass have long lifetimes, some comparable to the age of the Galaxy, /97/ $

2 504 MCWILLIAM and their envelopes have preserved much of their original chemical composition. These stars are useful because they are fossils containing information about the history of the evolution of chemical abundances in the Galaxy. After the Big Bang, the story of nucleogenesis is concerned mostly with the physics of stellar evolution and nucleosynthesis in stars, with how the environment dictated the kinds of stars that formed to enrich the Galactic gas, and with how the enriched gas mixed with the interstellar medium to form subsequent stellar generations (Hoyle 1954). We can try to understand these processes and chemical evolution from theoretical models, but the best way to learn about the history of the elements in the Galaxy is to look at the fossils. Time and space do not permit me to discuss in sufficient detail the many exciting developments that have occurred in the area of chemical evolution. Therefore, I restrict myself to areas most closely aligned with my research, which is usually concerned with high-resolution abundance analysis of stars in our Galaxy; in particular I do not discuss all the elements, or families of elements, and some elements may be conspicuous by their absence. Despite its obvious flaws, a good starting point for developing a mental picture of chemical evolution is the Simple one-zone model (e.g. Schmidt 1963, Searle & Sargent 1972, Pagel & Patchett 1975). The model assumes evolution in a closed system, with generations of stars born out of the interstellar gas (ISM). In each generation, a fraction of the gas is transformed into metals and returned to the ISM; the gas locked up in long-lived low-mass stars and stellar remnants no longer takes part in chemical evolution. Newly synthesized metals from each stellar generation are assumed to be instantaneously recycled back into the ISM and instantaneously mixed throughout the region; thus, in this model, metallicity always increases with time, and the region is perfectly homogeneous at all times. The ratio of mass of metals ejected to mass locked up, y, is a quantity commonly called the yield. The term yield has another meaning: Supernova (SN) nucleosynthesis theorists use it to refer to the mass of a particular element ejected in a SN model. The yield depends on the mass of metals ejected by stars (usually a function of mass) and the relative frequency of different mass stars born in a stellar generation (this is the initial mass function, or IMF). The mean IMF has been measured empirically (e.g. Scalo 1987) and over Galactic time appears to have been approximately constant; however, for individual molecular clouds, large deviations from the mean IMF occur. Another chemical evolution parameter is the star formation rate (SFR), which has been postulated to be proportional to some power of the gas density and the total mass density. In the Simple model, the SFR affects the time evolution of the metallicity but does not affect the final metallicity function of the system after the gas has been exhausted.

3 ABUNDANCE RATIOS 505 Given the yield, the metallicity function of long-lived stars for the Simple model is as follows: f (z) = y 1 exp( z/y) If evolution continues to gas exhaustion, then the Simple model predicts that the average mass fraction of metals of long-lived stars is equal to the yield, <z> = y. In principle the mean metal content of a stellar system can tell us about the yield. Because the yield is the ratio of mass of metals produced to the mass in low-mass stars per generation, it is sensitive to the IMF: An IMF skewed to high-mass stars would have a higher yield because more stars are massive enough to produce metals as SN, and there are fewer low-mass stars to lock away the gas. Abundance ratios can serve as a diagnostic of the IMF and SFR parameters and time scale for chemically evolving systems. Tinsley (1979) proposed that type Ia supernovae (SN Ia, resulting from mass accretion by a C-O white dwarf) are the major producers of iron in the Galaxy and that the SN Ia progenitors have longer lifetimes than the progenitors of type II supernovae (SN II, resulting from exploding massive stars), which are the source of Galactic oxygen; Tinsley argued that the time delay between SN II and SN Ia, of at least 10 8 years, is responsible for the enhanced [O/Fe] 1 ratios observed in halo stars. Theoretical predictions of SN II element yields show that [α/fe] (where α includes the elements O, Mg, Si, S, Ca, and Ti) increases with increasing progenitor mass (e.g. Woosley & Weaver 1995). In principle, the IMF of a stellar system could be inferred from the observed [α/fe] ratios. Note that if a stellar system is found to have a high average metallicity, and an IMF skewed to high-mass stars is responsible for increasing the yield, then the composition should reflect an increased [α/fe] ratio that is due to the increased [α/fe] from high mass SN II. In fact, this idea was used by Matteucci & Brocato (1990) to explain the putative high metallicity of the Galactic bulge, with the prediction that [α/fe] is enhanced in the bulge. The [α/fe] ratio is also sensitive to the SFR in Tinsley s model: If the SFR is high, then the gas will reach higher [Fe/H] before the first SN Ia occur, and the position of the knee in the [α/fe] versus [Fe/H] diagram (Figure 1) will be at a higher [Fe/H]. Also, because the knee marks the time of the first SN Ia, then the formation time scale of a stellar system can be estimated by noting the fraction of stars with [Fe/H] below this point. Another potentially useful diagnostic of the [O/Fe] ratio was pointed out by Wyse & Gilmore (1991): In a star-burst system, the O/Fe ratio of the gas is 1 [A/B] refers to an abundance ratio in log 10 solar units, where A and B represent the number densities of two elements: [A/B] = log 10 (A/B) * log 10 (A/B). Note that ɛ(m) = log 10 (M/H).

4 506 MCWILLIAM Figure 1 A schematic diagram of the trend of α-element abundance with metallicity. Increased initial mass function and star formation rate affect the trend in the directions indicated. The knee in the diagram is thought to be due to the onset of type Ia supernovae (SN Ia). initially above solar owing to nucleosynthesis by SN II, but as time continues after the burst (with no new star formation) the SN II diminish, only SN Ia enrich the gas; ultimately subsolar [O/Fe] ratios occur. Wyse & Gilmore (1991) claimed that the composition of the LMC is fit by this model. Elements like C, O, and those in the iron-peak, thought to be produced in stars from the original hydrogen, are sometimes labeled as primary. The label secondary is reserved for elements thought to be produced from preexisting seed nuclei, such as N and s-process heavy elements. The abundance of a primary element is expected to increase in proportion to the metallicity, thus [M/Fe] is approximately constant. For a secondary element, [M/Fe] is expected to increase linearly with [Fe/H] because the yield is proportional to the abundance of preexisting seed nuclei. One difficulty is that N and the s- process elements (both secondary) do not show the expected dependence on metallicity. THE SOLAR IRON ABUNDANCE It is sobering, and somewhat embarrassing, that the solar iron abundance is in dispute at the level of 0.15 dex. This discrepancy comes in spite of the fact that more than 2000 solar iron lines, with reasonably accurate gf values, are available for abundance analysis; that the solar spectrum is measured with much higher S/N and dispersion than for any other star; that LTE corrections to Fe I abundance are small, at only dex (Holweger et al 1991); and

5 ABUNDANCE RATIOS 507 that both theoretical and empirical solar model atmospheres are available, with parameters known more precisely than for any other star. Anders & Grevesse (1989) reviewed published meteoritic and solar photospheric abundances for all available elements and found ɛ(fe) = 7.51 ± 0.01 for meteorites and ɛ(fe) = 7.67 from the solar abundance analysis of Blackwell et al (1984, 1986), which was a notable increase from the earlier photospheric value of 7.50 ± 0.08 favored by Ross & Aller (1976). Blackwell et al s work utilized the Oxford group gf values for Fe I lines, which are known to be of high accuracy. Pauls et al (1990) found ɛ(fe) = 7.66 from Fe II lines, but Holweger et al (1990), also using Fe II lines, found ɛ(fe) = Biémont et al (1991) measured the solar iron abundance of 7.54 ± 0.03 with a larger sample of Fe II lines. Holweger et al (1991) found 7.50 ± 0.07 based on gf values for Fe I lines measured by Bard et al (1991). Two recent papers are characteristic of the conflicting solar iron abundance: those by Holweger et al (1995) and Blackwell et al (1995). Blackwell et al (1995) employed Oxford gf values and the Holweger & Müller (1974) solar atmosphere and found ɛ(fe) = 7.64 ± 0.03 from the Fe I lines; although the Fe II line results indicated ɛ(fe) = 7.53 dex. When Blackwell et al (1995) computed iron abundances from the Kurucz (1992 unpublished) solar model, Fe I and Fe II lines gave better agreement, at 7.57 and 7.54 dex, respectively, but they claimed that the Kurucz model results are not valid because the solar limb darkening is not reproduced by the model. Blackwell et al (1995) concluded that neither the empirical Holweger-Müller model, nor the Kurucz theoretical model atmosphere, is adequate for measuring the solar iron abundance. Holweger et al (1995) contested Blackwell et al s (1995) claim and argued that Fe I lines analyzed with the Holweger-Müller model give ɛ(fe) = 7.48 ± 0.05, or 7.51 with the 0.03-dex non local thermodynamic equilibrium (non LTE) correction. Holweger et al (1995) found the same low solar iron abundance from both Fe I and Fe II lines in their analysis. Lambert et al (1995a) found that the gf values of lines common to results of both Holweger et al (1995) and Blackwell et al (1995) had zero average difference, which suggests that gf values are not the source of the abundance difference. They attributed the difference mostly to variations in the measured equivalent widths and damping constants. Another low value of the solar iron abundance was found by Milford et al (1994), who found ɛ(fe) = 7.54 ± 0.05 with the Holweger-Müller solar model and new gf values, from weak Fe I lines that are not sensitive to uncertainties in damping constants or microturbulent velocity. Kostik et al (1996) attempted to resolve the differences between Blackwell et al (1995) and Holweger et al (1995). Kostik et al found that the Blackwell

6 508 MCWILLIAM et al s equivalent widths are systematically higher than Holweger et al s values; remeasurement by Kostik et al favored the Holweger et al values. Kostik et al also found suspicious trends in the gf values of the Holweger et al (1995) study, and they agree with Grevesse & Noels (1993) that the spread in iron abundance is dominated by uncertainties in the gf values. They also noted that uncertainties in the microturbulent velocity and collisional damping constants are extremely important to the adopted value. Kostik et al provide a best estimate of the solar iron abundance of 7.62 ± 0.04, which favors the high solar iron abundance; although little weight was placed on the significance of this result. Anstee et al (1997) measured the solar iron abundance from profile-matching 26 strong Fe I lines, using accurate laboratory collision-damping constants and gf values. They found ɛ(fe) = 7.51 ± 0.01 in complete agreement with the meteoritic iron abundance of Anders & Grevesse (1989), independent of nonthermal motions in the photosphere. Anstee et al traced the discrepancies between previous studies to the use of different atomic data, measured equivalent widths, and assumed microturbulent velocity. It now seems that the weight of the evidence favors the low value of the solar iron abundance, and the issue may finally be settled; however, this statement has been made before... SUPER METAL-RICH STARS The existence of super metal-rich (SMR) stars was first claimed by Spinrad & Taylor (1969), based on low-resolution spectra. The term SMR is generally meant to signify that a star is more metal-rich than the sun by an amount that cannot be explained as simple measurement error. The existence of SMR stars is, historically, a controversial subject; the main question is whether SMR stars are really metal-rich or just appear so because of some kind of measurement dispersion or systematic error. Perhaps the notion of SMR stars became more acceptable with claims that the Galactic bulge red giant stars are on average more metal-rich than the sun (e.g. Whitford & Rich 1983, Frogel & Whitford 1987, Rich 1988). McWilliam & Rich (1994) showed that the average bulge [Fe/H] is the same as in the solar neighborhood, but that the most metal-rich bulge giant, BW IV-167, at [Fe/H] =+0.44 is almost identical to µ Leo, a metal-rich disk giant. Taylor (1996) has reviewed abundance estimates for SMR stars, including low- and high-resolution results, and concluded that true SMR stars do not exist. Given the controversy and the potential significance for chemical evolution, it seems important to establish whether any firm cases of SMR stars exist at all. In the Galactic disk, the most well-studied SMR candidate is the K giant star µ Leo. High-resolution high-s/n model atmosphere abundance analyses of µ

7 ABUNDANCE RATIOS 509 Leo have been performed by several groups: Gustafsson et al (1974), Branch et al (1978), Brown et al (1989), Gratton & Sneden (1990), McWilliam & Rich (1994), and Castro et al (1996) all found values near [Fe/H] =+0.45 for a solar scale of ɛ(fe) = 7.52; on the other hand, Lambert & Ries (1981), McWilliam (1990), and Luck & Challener (1995) found [Fe/H] from +0.1to+0.2dex. Metal-rich stars in the McWilliam (1990) study (e.g. µ Leo) were affected by two systematic problems: CN blanketing depressed most of the µ Leo continuum regions in the two small 100-Å portions of the spectrum observed (found by McWilliam & Rich 1994), which resulted in smaller equivalent widths; second, McWilliam (1990) did not have access to metal-rich model atmospheres, which caused underestimation of the [Fe/H] for metal-rich stars ( 0.1-dex underestimate for µ Leo). Both of these effects decreased the measured µ Leo [Fe/H] in the McWilliam (1990) work; accounting for the model atmosphere correction alone would increase [Fe/H] to dex. The Luck & Challener (1995) study concluded that their sample of stronglined stars showed only small iron abundance enhancements at [Fe/H] +0.1 dex; in the case of µ Leo they found [Fe/H] =+0.20 dex. Luck & Challener (1995) chose not to use a SMR model atmosphere for µ Leo, thus artificially lowering the computed [Fe/H] by 0.08 dex (Castro et al 1996). Castro et al (1996) showed that the low Luck & Challener [Fe/H] must result from differences in analysis because of the good agreement between equivalent widths of lines in common. Furthermore, Luck & Challener confused the [A/H] = 0.0 of the Bell et al (1976) atmosphere grid with a solar iron abundance of ɛ(fe) = 7.67 (from Anders & Grevesse 1989), whereas the models were actually calculated with ɛ(fe) = Castro et al noted that when these two problems are taken into account the Luck & Challener result for µ Leo becomes [Fe/H] =+0.43, assuming the solar ɛ(fe) = Thus the most recent high resolution abundance studies of µ Leo that are discordant with the notion of [Fe/H] =+0.45 can be readily resolved, and it appears that there is a convergence of the µ Leo iron abundance near [Fe/H] =+0.45 dex with the assumed low value for the solar iron abundance. I do not have an explanation for the Lambert & Ries (1981) low [Fe/H], although it seems possible that the heavy line blanketing and limited spectral coverage may have affected the continuum placement. Studies with the highest S/N spectra, and the most detailed abundance analyses (e.g. Gratton & Sneden 1990, Branch et al 1978, Castro et al 1996, McWilliam & Rich 1994), consistently find [Fe/H] +0.4 dex for µ Leo. In conclusion, the high dispersion abundance analyses confirm at least one case of super metallicity. High-resolution abundance analyses of SMR stars have also been carried out by Edvardsson et al (1993), who found F dwarf stars up to [Fe/H] =+0.26

8 510 MCWILLIAM dex; Feltzing (1995), who extended the Edvardsson sample to find stars between [Fe/H] of 0.08 and dex; and Castro et al (1997), who studied a subset of the sample identified by Grenon (1989) and found [Fe/H] ranging from to dex. McWilliam & Rich (1994) found two SMR Galactic bulge giants, BW IV-167 and BW IV-025, with [Fe/H] of and dex, respectively. It appears that high-resolution abundance studies do find SMR stars with [Fe/H] up to approximately dex. OBSERVED METALLICITY DISTRIBUTION FUNCTION In this section, I discuss some implications and uses of the most basic chemical composition information, namely metallicity. The word metallicity has more than one meaning: The precise definition is that metallicity is the mass fraction of all elements heavier than helium, denoted by the symbol Z; this is not always practical for observers because information usually does not exist for all elements. For observational stellar astronomy, metallicity is more often used to refer to the iron abundance. Unless explicitly stated the word metallicity used here refers to [Fe/H], the logarithmic iron abundance relative to the solar value. The Disk Because the main-sequence lifetimes of G and F dwarfs are comparable to the age of the Galaxy, all the G dwarfs ever born are assumed to still exist (although see discussion of metallicity-dependent lifetimes by Bazan & Mathews 1990), and so these stars can provide a complete picture of Galactic chemical evolution. Early studies of the metallicity distribution of G dwarfs, within about 25 pc of the sun (vandenbergh 1962, Schmidt 1963, Pagel & Patchett 1975), showed that there is a deficit of metal-poor stars relative to the prediction of the Simple model; this is the well-known G-dwarf problem. The metallicities of these early studies were based on UV excesses (see Wallerstein & Carlson 1960, Sandage 1969), which are accurate to approximately 1σ = 0.25 dex (Pagel & Patchett 1975); although Norris & Ryan (1989) claim uncertainties of ±0.45 dex. The observed metallicity distributions contain biases that must be taken into account in order to obtain the true metallicity function (e.g. see Sommer-Larsen 1991 and Pagel 1989). Many possible explanations were presented to account for the G-dwarf problem (e.g. Audouze & Tinsley 1976), but infall of metal-poor gas onto the disk was the most favored solution. To fit the observed metallicity function by this scheme, the original disk was at most 5% of the present disk mass (Pagel 1989), with mass infall occurring over several billion years. Variants of the Simple model exist that include gas infall in various ways (e.g. Larson 1974, 1976,

9 ABUNDANCE RATIOS 511 Lynden-Bell 1975, Clayton 1985, 1988, Pagel 1989). All of these models predict a strict age-metallicity relation (AMR) with no abundance dispersion. In these models, the halo could not have been responsible for the bulk of the gas infall because the present-day luminous halo mass is only a few percent of the disk (Sandage 1987, Pagel 1989); a metallicity function of the disk+halo still suffers a paucity of metal-poor stars relative to the simple model (e.g. Worthey 1996). Tosi (1988) showed that infall of gas with metallicity 0.1 Z provides as effective an explanation of the observed disk metallicity distribution function as infall of zero metallicity gas; however, infalling gas with Z = 0.4 Z is excluded by observations. A number of studies over the last decade and a half have combined star count and kinematic information with metallicities estimated from UV excesses (e.g. Sandage & Fouts 1987), ubvyβ photometry (e.g. Nissen & Schuster 1991), and low S/N spectra (e.g. Carney et al 1987, Jones et al 1995). The assembled databases have been used to imply the existence of various Galactic populations. For example, the thick disk of Gilmore & Reid (1983) is characterized by scale height of 1.3 pc, mean [Fe/H] 0.6dex, and dispersion 0.3 dex (Gilmore & Reid 1983, Gilmore 1984, Gilmore & Wyse 1985, Wyse & Gilmore 1986), with no apparent metallicity gradient. Wyse & Gilmore (1995) conclude that the data are best fit by overlapping thick and thin disks; the thick disk has a mean metallicity of [Fe/H] 0.7dex, ranging from 0.2to 1.4dex. A low metallicity tail, extending down to [Fe/H] 2to 3, was claimed by Norris & Ryan (1991), Beers & Sommer-Larsen (1995), and Pagel & Tautvaisiene (1995). Typical star count models yield thick disk to thin disk ratios of a few percent (e.g. Majewski 1993). The thin disk metallicity peaks near [Fe/H] = 0.25 dex, ranging from +0.2to 0.8 dex (Wyse & Gilmore 1995). The Halo The Galactic halo does not appear to suffer from a severe G-dwarf problem (Laird et al 1988, Pagel 1989, Beers et al 1992). The halo metallicity ranges from 4 dex to just below the solar value, with a mean of 1.6 (Laird 1988, Hartwick 1976); Hartwick (1976) noted that this low metallicity suggested that either the halo yield was much lower than in the disk or that gas was removed from halo star formation (e.g. Ostriker & Thuan 1975). The favored model is that the halo lost its gas before chemical evolution could go to completion. Carney et al (1990) and Wyse & Gilmore (1992) suggested that the missing spheroid mass fell to the center of the Galaxy and contributed most of the bulge mass, based on angular momentum considerations. Whether or not there is a minimum metallicity level, below which stars do not exist, has been debated for at least 20 years. Hartquist & Cameron (1977) predicted that there was an era of pregalactic nucleosynthesis by very massive

10 512 MCWILLIAM zero metallicity objects; as a result, the Galactic halo would have formed with a non zero metal content. Bond (1981) and Cayrel (1987) claimed that there is a paucity of stars below [Fe/H] 3 relative to a Simple one-zone model of chemical evolution; this was attributed to a reduced efficiency of forming low-mass stars at low metallicity. Indeed several theoretical investigations (e.g. Kahn 1974, Wolfire & Cassinelli 1987, Yoshii & Saio 1986, and Uehara 1996) have predicted that at low metallicity the IMF is skewed to high-mass stars. Contrary to Bond s suggestion, the huge increase in the number of known metal-poor halo stars (e.g. Beers et al 1985, 1992) led to agreement between the observed metallicity function and predictions from modified Simple models (Beers et al 1985, 1992, Laird et al 1988, Ryan & Norris 1991) down to the lowest measurable abundance, consistent with no metallicity dependence of the IMF. Audouze & Silk (1995) claimed that there is a lower limit to the metallicity that can form stars, based on predictions concerning the amount of material that can dilute and cool SN ejecta; they estimated the lower limit to be approximately [Fe/H] 4. The most metal-poor star presently known is CD , at [Fe/H] = 4.01 (McWilliam et al 1995a,b), although it only narrowly beats CS for the record, at [Fe/H] = This iron abundance for CD is supported by Gratton & Sneden (1988), who found [Fe/H] = 3.97, but it is higher than the metallicity of Bessell & Norris (1984), who found [Fe/H] = 4.5. Norris et al (1993) also analyzed stars from the list of Beers et al (1992), one of which was CS , with a measured [Fe/H] = McWilliam et al (1995a) found [Fe/H] = 3.79 for this star and explained the difference as due to systematic analysis effects of 0.4 dex; if applied to the Bessell & Norris (1984) result, the same zero point would bring all three analyses into agreement at [Fe/H] = 4.0 for CD Thus, despite the heroic effort by George Preston of searching for metal-poor stars by visually inspecting over one million objective prism spectra (Beers et al 1985), the honor of the most metal-poor star known in the Galaxy still belongs to CD The Bulge Measurement of the metallicity of Galactic bulge stars has been somewhat controversial in the last 15 years. Early bulge metallicity studies focused on stars in Baade s window, at Galactic latitude 4. Initial low-resolution studies of 21 bulge giants by Whitford & Rich (1983) suggested that most of the bulge stars are super metal-rich. Frogel & Whitford (1987) amassed photometric and spectral-type data for a large number of bulge giants. They found that bulge M giants have stronger TiO

11 ABUNDANCE RATIOS 513 and CO bands than solar neighborhood M giants consistent with a metal-rich bulge. Rich (1988) measured low-resolution indices of strong lines (Mg b and Fraunhofer Fe I lines) in 88 bulge giants in Baade s window and several bright standards. Calibration of the indices suggested a range of [Fe/H], from 1.0 to +0.8 dex for the bulge, with a mean value twice the solar value. Terndrup et al (1991) found a mean bulge metallicity of +0.3 dex for M giant stars in Baade s Window, based on R=1000 spectrophotometry, which confirmed earlier results. Geisler & Friel (1992) used Washington photometry to measure the metallicity of 314 red giants in the Galactic bulge, through Baade s window. They found the mean [Fe/H] = ± 0.15 dex, in good agreement with Rich (1988). They also found a high frequency of metal-poor stars, consistent with that expected from a simple closed box model, as found by Rich (1990). Rich (1990) showed that the Galactic bulge contains a higher frequency of metal-poor stars than the solar neighborhood. In fact, the bulge metallicity function does not exhibit the G-dwarf problem. This is perhaps somewhat surprising because the bulge must be the final repository of infalling material (for example, the Sagittarius dwarf found by Ibata et al 1994). It may be that most of the infall occurred very rapidly, or that material that fell into the bulge, such as a dwarf galaxy, was stripped of its gas before reaching the bulge. With the apparent convergence of different methods used to measure the bulge metallicity, it was a surprise that the first high-dispersion model atmosphere abundance analysis of bulge stars (McWilliam & Rich 1994) found that the bulge is slightly iron-poor relative to the solar neighborhood. McWilliam & Rich (1994) computed [Fe/H] for 11 bulge red giants, covering the full metallicity range, which had previously been measured by Rich (1988). A correlation of [Fe/H] values of McWilliam & Rich (1994) with those of Rich (1988) showed that Rich (1988) systematically overestimated the [Fe/H] of the most metalrich stars. A regression relation between McWilliam & Rich s (1994) and Rich s (1988) [Fe/H] results was used to compute corrected [Fe/H] (Rich 1988) for the full sample of 88 stars. Rich s (1988) [Fe/H] values corrected in this way have a mean of 0.25 dex, slightly below the mean value of 0.17 dex for solar neighborhood red giants (McWilliam 1990). The corrected bulge metallicity function still shows the excess of metal-poor stars relative to the solar neighborhood noted by Rich (1990). McWilliam & Rich (1994) also found unusually high [Mg/Fe] and [Ti/Fe] ratios in the bulge stars, which might explain why previous investigators found high average metallicities. Subsequent model atmosphere abundance analyses of two stars in the McWilliam & Rich (1994) sample (Castro et al 1996; A McWilliam, RC Peterson, DM Terndrup & RM Rich, in preparation) confirmed the McWilliam

12 514 MCWILLIAM & Rich (1994) [Fe/H] results. The low [Fe/H] of bulge stars in Baade s window found by McWilliam & Rich (1994) was supported by later low-resolution studies; for example, the analysis of low-resolution spectra of 400 bulge giants by Terndrup et al (1995) and Sadler et al (1996) found a low mean [Fe/H] 0.1 dex. AGE-METALLICITY RELATION The existence of an age-metallicity relation (AMR) in the disk is an important issue for developing chemical evolution models. There is currently some uncertainty whether an AMR exists: Studies of open cluster metallicities and ages (e.g. Arp 1962, Geisler 1987, Geisler et al 1992, Friel & Janes 1993) have resulted in the conclusion by some that there is no AMR in the Galactic disk (see the review by Friel 1995). The main factor in determining open cluster metallicity appears to be galactocentric radius (e.g. Geisler et al 1992). It is also clear that there is a large scatter in metallicity at any given age in the disk: The dispersion in the age-metallicity diagram is exemplified by the presence of very old open clusters with metallicities near or above the solar value. The open cluster NGC 188 has historically been used to illustrate this point (e.g. Eggen & Sandage 1969); but the most clear-cut modern case is NGC 6791, which is more metal-rich than the sun, with [Fe/H] +0.2 to +0.3 dex (Peterson & Greene 1995, Montgomery et al 1994), but very old at years (Montgomery et al 1994, Tripicco et al 1995). The conclusion against an AMR is at odds with claims based on studies of field stars. For example, Twarog (1980), Meusinger et al (1991), and Jønch- Sørensen (1995) all employed uvbyβ photometry and found a trend of decreasing metallicity with increasing stellar age. Edvardsson et al (1993) used spectroscopic abundance analysis to determine [Fe/H] and uvbyβ photometry for the ages. They found an AMR consistent with the results of Twarog (1980) and Meusinger et al (1991) but with a considerable scatter about the mean trend (Figure 2). The Jønch-Sørensen data indicated a similar AMR slope and scatter as Edvardsson et al s data. The age-metallicity diagram from these studies (e.g. Figure 2) show a lower envelope to the observed metallicity of stars that increases with Galactic time; in particular, no young stars with [Fe/H] 1 have been found in the solar neighborhood (although low metallicity stars at large galactocentric radii are known; e.g. Geisler 1987, Geisler et al 1992). The large scatter in metallicity at all ages is the one consistent conclusion common to the age-metallicity diagrams for both the field stars and open clusters. François & Matteucci (1993) suggested that the scatter could be due to orbital diffusion; however, Edvardsson et al (1993) showed that this is not enough to reduce the observed scatter in the age-metallicity diagram.

13 ABUNDANCE RATIOS 515 Figure 2 The age-metallicity relation for the solar neighborhood, from the data of Edvardsson et al (1993). The sample is limited to galactocentric radius 7.7 R m 9.3 kpc, maximum height above the plane Z max 0.26 kpc, and eccentricity e The position of the Sun is indicated. It is clear that certain biases occur in samples of field stars that could conspire to create an apparent AMR, even if none exists (Knude 1990, Grenon 1987); indeed, Friel (1995) states that the age-metallicity trends seen by Twarog (1980) are the result of these selection effects. However, Twarog was aware of the selection biases and pointedly went to great effort to avoid them. Jønch-Sørensen (1995) estimated an upper limit to the number of metal-poor young stars and claimed that the selection bias against metal-poor young stars could not account for the apparent AMR. Edvardsson et al (1993) made a correction for a metallicity bias, but the AMR was still present. Obviously a definitive resolution to the existence or absence of a mean AMR in field stars would be extremely valuable. If age and metallicity data for the halo are added to Figure 2, as done by Eggen & Sandage (1969), a strong AMR would result; however, the validity of combining these two populations is not certain. The large range of metallicities present for all ages suggests that chemical enrichment up to solar metallicity can occur on rapid time scales ( years) and that the disk has been chemically inhomogeneous throughout its development. The dispersion in the AMR at the solar circle (as seen in the Edvardsson et al 1993 study) shows that the composition of the Galactic disk did not evolve homogeneously. Traditional chemical evolution models, for example those of Lynden-Bell (1975), Larson (1976), Matteucci & François (1989), Pagel (1989), Sommer-Larsen (1991), and Pagel & Tautvaisiene (1995), cannot account for the observed AMR dispersion because they all assume instantaneous

14 516 MCWILLIAM mixing of recycled gas and a homogeneous steady infall; as a consequence chemical homogeneity is preserved at all times. Reeves (1972) suggested that significant spatial inhomogeneities in elemental abundances could occur as a result of self-enrichment of star-forming regions by SN events. However, Edmunds (1975) investigated this possibility and concluded that the Galactic disk is well mixed. White & Audouze (1983) developed analytical expressions that extended the standard chemical evolution model of Lynden-Bell (1975) to the case of inhomogeneous steady-state evolution. Two important mixing parameters dictated the inhomogeneity: (a) the mean mass of disk material mixed with a unit mass of enriched material from star formation events and (b) the mean mass of disk material mixed with a unit mass of infalling gas. Recent models of Galactic evolution attempted to describe inhomogeneous chemical evolution: Pilyugin & Edmunds (1996a,b) and Raiteri et al (1996). Both studies adopt the Twarog (1980) AMR and the dispersion about this relation indicated by Edvardsson et al (1993). Pilyugin & Edmunds (1996b) considered inhomogeneity by two mechanisms. In the first approach, self-enrichment of gas in star forming regions (H II regions) for years is permitted, after which time the gas is instantaneously mixed with the ambient disk gas. This approximates a star-forming region in which SN ejecta enrich the region with metals until the energy input from SN is sufficient to disrupt the cloud in years, followed by mixing with the disk in 10 8 years. Justification for this assumption comes from Cunha & Lambert (1992, 1994), who showed that self enrichment in the Orion association has occurred in years, based on enhancements in O and Si abundances as a function of age of the Orion subgroups. Self enrichment of the H II regions gave a satisfactory fit to the dispersion in the oxygen abundance with time, but it was incapable of reproducing the observed dispersion in Fe abundance. The difficulty in reproducing the Fe dispersion was caused by the fact that Fe is produced mainly in SN Ia, whose progenitor lifetimes are thought to be years, well in excess of the self-enrichment time scale. Pilyugin & Edmunds (1996b) suggest that selfenrichment of H II regions results in larger dispersion for oxygen abundances (SN II progenitors with short lifetimes) than iron abundances versus age. They concluded that the large observed dispersion for both O and Fe implicates another source of inhomogeneity. Pilyugin & Edmunds (1996b) suggested that episodic gas infall could account for the large dispersions in the AMR for both Fe and O. If infalling gas fell onto the disk in a nonuniform fashion (both temporally and spatially), then disk gas could reach solar metallicity followed by substantial dilution to lower metallicities. Stars formed over such a cycle would exhibit equal Fe and O

15 ABUNDANCE RATIOS 517 dispersion in the AMR because dilution affects all species equally. If this is the case, then the infalling gas cannot be pure hydrogen; otherwise the dilution would preserve solar abundance ratios even near [Fe/H] = 1, which is not observed. The gas would need to be of halo composition, with [Fe/H] 1, to avoid the problem of solar ratios in low metallicity disk stars. Raiteri et al (1996) have developed N-body/hydrodynamical simulations of Galactic chemical evolution. The method seems very promising and does produce an AMR similar to Twarog s (1980) with a large metallicity dispersion; it also predicts significant dispersion in the [O/Fe] ratio at all metallicities, which provides a basis for testing the model. There are some problems, however, such as a very high frequency of low metallicity stars. ABUNDANCE TRENDS WITH METALLICITY Alpha Elements Enhancements of α elements in metal-poor stars were first identified by Aller & Greenstein (1960) and more firmly established by Wallerstein (1962), who found excesses of Mg, Si, Ca, and Ti relative to Fe. A corresponding enhancement for oxygen was first discovered by Conti et al (1967). The work of Clegg et al (1981) and François (1987, 1988) showed that S is also overabundant in metal-poor stars. These enhancements increase linearly with decreasing metallicity, reaching a factor of two above the solar [α/fe] ratios at [Fe/H] near 1; below [Fe/H] = 1 the enhancements are approximately constant. Figure 3a shows the general trend of [O/Fe] with [Fe/H]. It is important to emphasize that α element is simply a convenient phrase used to signify the observation that some even-z elements (O, Mg, Si, S, Ca, and Ti) are overabundant relative to iron at low metallicity, and it does not signify that these are all products of a single nuclear reaction chain that occurs in the same astrophysical environment. As mentioned in the introduction Tinsley (1979) suggested that the [α/fe] trend with [Fe/H] is due to the time delay between SN II, which produce α elements and iron-peak elements (e.g. Arnett 1978, Woosley & Weaver 1995), and SN Ia, which yield mostly iron-peak with little α element production (e.g. Nomoto et al 1984, Thielmann et al 1986). Thus, after the delay for the onset of SN Ia, the [α/fe] ratio declines from the SN II value. The SN Ia time scale is an important consideration for this model. Iben & Tutukov (1984) favor a mechanism with mass transfer during the merging of a CO+CO white dwarf binary system; time scales for SN Ia from this model range from 10 8 to years, depending on progenitor masses and mass transfer parameters. Smecker-Hane & Wyse (1992) obtained estimates for the first SN Ia of 10 8 years. Other explanations for the α-element trend have been put forward: Maeder (1991) suggested that exploding Wolf-Rayet stars (type Ib supernovae, SN Ib)

16 518 MCWILLIAM Figure 3 The trend of oxygen abundance with metallicity. The favored trend is shown in (a), a compilation of [O I] results: crosses from disk data of Edvardsson et al (1993), filled squares from Spite & Spite (1991), filled circles from Barbuy (1988), open triangles from Kraft et al (1992) and Sneden et al (1991), open squares from Shetrone (1996a). (b) shows results from the O I triplet: crosses (Abia & Rebolo 1989) and filled triangles (Tomkin et al 1992); low S/N results from ultraviolet OH lines are indicated by open squares (Nissen et al 1994) and open triangles (Bessell et al 1991). Note the difference in the scale of the ordinate between (a) and (b).

17 ABUNDANCE RATIOS 519 might be responsible for the observed α-element abundance trend. Wolf-Rayet stars are the bare cores of massive stars that have lost their outer envelopes through copious stellar winds. The radiatively driven winds are metallicitydependent, producing significant numbers of Wolf-Rayet stars above [Fe/H] 1. The chemical yields depend on the mass-loss rates: At high metallicity the strong winds remove much of the helium before it is further transformed into heavy elements. Edmunds et al (1991) suggested that metallicity-dependent element yields could be the source of the α-element abundance trend and predicted that SMR stars should possess subsolar [α/fe] ratios. The theoretical element yields from SN II (e.g. Woosley & Weaver 1995) do not show such a metallicity dependence; however, some star formation theories have predicted a metallicity-dependent IMF (e.g. Kahn 1974, Yoshii & Saio 1986), which might conceivably result in a steady increase of the SN Ia/SN II ratio with increasing metallicity and thereby account for the observed α-element trend. Disk Alpha Elements Studies of disk dwarf stars by several workers (e.g. Clegg et al 1981, Tomkin et al 1985, François 1986, Gratton & Sneden 1987, Edvardsson et al 1993) confirmed the trend of increasing [α/fe] with decreasing [Fe/H] in the Galactic disk, as established by the analysis of G dwarfs by Wallerstein (1962); typically [α/fe] +0.4 at [Fe/H] 1.0. The data of Tomkin et al (1986) and Edvardsson et al (1993) show that for Mg, Ca, and Si, there is a plateau at [α/fe]= 0.0 above [Fe/H] 0.2 dex (see Figure 3a). This plateau suggests a transition from one kind of chemical evolution environment to another, which is consistent with the idea that above [Fe/H] = 0.2, the ratio of SN Ia/SN II had reached a constant value. Edvardsson et al (1993) found that when the disk stars are separated into bins of mean galactocentric radius, R m, the α-element enhancements are seen to be maintained to higher [Fe/H] at small R m (see Figure 4). In Tinsley s picture of SN Ia and SN II this suggests that enrichment by SN II occurred to higher [Fe/H] in the inner disk than in the outer disk, before the first SN Ia occurred, in agreement with models of the disk that predict higher SFR in the inner disk than in the outer regions (e.g. Larson 1976, Matteucci & François 1989). Edvardsson et al s results also indicate that at the solar circle, old stars seem to show a distinctly different [α/fe] trend than young stars, and this suggests that the SFR increased with time in the disk. There is a hint that the inner disk stars of the Edvardsson et al sample show a bimodal [α/fe] ratio, rather than a slope with [Fe/H]. In order to reduce scatter in the trend with metallicity, α-element abundances have often been averaged; Lambert s (1987) review popularized the mean relation between [α/fe] and [Fe/H]. The work of Edvardsson et al (1993) indicated

18 520 MCWILLIAM (a) R<7kpc (b) (c) Figure 4 (continued )

19 ABUNDANCE RATIOS 521 (d) Figure 4 The run of [α/fe], computed from [(Mg+Si+Ca+Ti)/Fe], versus iron abundance for four ranges in galactocentric radius, from the data of Edvardsson et al (1993). In the inner disk, R m 7 kpc, [α/fe] is higher than the mean trend (indicated by the solid line), while the outer disk shows an [α/fe] deficiency. The trends at large and small R m seem to show a bimodal appearance, rather than the shift indicated in Figure 1 for different SFR. that the trends are not the same for all α elements: Ca and Si abundances correlate very well, but both Mg and Ti are systematically over-enhanced relative to Ca and Si. These observations of subtle α-element trends in the disk stars are similar to, but less extreme than, the enhanced Mg and Ti abundances found for Galactic bulge stars by McWilliam & Rich (1994). Nissen & Edvardsson (1992) found a somewhat steeper decline in [O/Fe] with [Fe/H] than other α elements from Edvardsson et al (1993). If these differences within the α element family withstand further scrutiny it shows that the α elements are not made in a single process but are produced in different amounts by different SN. Cunha & Lambert (1992, 1994) studied the chemical composition of B stars in various subgroups of the Orion association ([Fe/H] 0.05) and found evidence for self-contamination of the association by nucleosynthesis products from SN II. In particular the subgroups show an abundance spread of 0.3 dex for O, correlated with Si abundance, but no dispersion larger than the measurement uncertainties could be found for Fe, C, and N. This pattern of abundance enhancement is consistent with self-enrichment of the gas by SN II only. Additional support for this idea includes the spatial correlation of the O-Si rich stars and the fact that the most O-Si rich stars are found only in the youngest subgroup of the association. The time lag between the oldest and youngest subgroups is years (Blaauw 1991), which is comparable to the lifetime of the massive stars. Thus, the massive stars had enough time to explode as SN and enrich the molecular cloud, but the time scale was too short

20 522 MCWILLIAM Figure 5 Trends of α-element abundances in the Galactic bulge, from McWilliam & Rich (1994). Filled triangles indicate the average [(Mg+Ti)/Fe] and open boxes indicate the average [(Si+Ca)/Fe]. For Si and Ca the trends follow the solar neighborhood relation (solid line), whereas the Mg and Ti abundances are enhanced by 0.4 dex for most stars, similar to the halo values. to permit any pollution by SN Ia. If the same enrichment observed by Cunha & Lambert (1994) occurred in a similar cloud of zero-metal gas, the metallicity of the final generation would be approximately [Fe/H] = 0.8 dex. As demonstrated by Cunha & Lambert (1992, 1994), chemical abundance studies of star-forming regions are a particularly useful way to study basic processes in chemical evolution and SN nucleosynthesis. Bulge Alpha Elements To date, the only extant detailed abundance analyses of α elements for Galactic bulge stars are by McWilliam & Rich (1994) and A McWilliam, A Tomaney & RM Rich (in preparation). McWilliam & Rich (1994) found that Mg and Ti are enhanced by +0.4 dex in almost all bulge stars, even at solar [Fe/H]; however, the abundances of Ca and Si appear to follow the normal trend of α/fe ratio with [Fe/H] (see Figure 5). Some overlap exists between the chemical properties of the McWilliam & Rich (1994) bulge giant sample and the disk F dwarfs of Edvardsson et al (1993): In general, the disk results (Edvardsson et al 1993) show that Mg and Ti are slightly enhanced relative to Si and Ca, which is similar to, but less extreme than, the +0.4-dex enhancements of Mg and Ti in the bulge. Edvardsson et al (1993) identified a subgroup of stars with 0.1-dex enhancements of Na, Mg, and Al; these are conceivably related to the bulge giants, which have large Mg and

21 ABUNDANCE RATIOS 523 Figure 6 Production factors from models of SN II by Woosley & Weaver (1995). Ejected element abundances for various progenitor masses are indicated by connected symbols; O and Mg are produced in large quantitiesat high mass ( 35 M ) but not in the lower mass (15 25 M ) SN, which are responsible for most of the Si and Ca production. None of the models give significant enhancements of Ti relative to Fe, contrary to observations of stars in the Galactic bulge and halo. Note that production factor is defined as the ratio of the mass fraction of an isotope in the SN ejecta, divided by its corresponding mass fraction in the Sun. The mass of the progenitor making the indicated elements is given in the key in the upper right. Al enhancements. The bulge [O/Fe] ratio is not well constrained: The extant data are insufficient to determine whether oxygen behaves like Mg and Ti or Si and Ca. However, any oxygen enhancement in the bulge must be less than +0.5 dex (A McWilliam, A Tomaney & RM Rich, in preparation). The unusual mixture of α-element abundances in the bulge is evidence that α elements are made in different proportions by different SN; i.e. there are different flavors of SN with different α-element yields. This conclusion is borne out by predicted α-element yields (e.g. Woosley & Weaver 1995), as shown in Figure 6. Figure 6 illustrates that enhanced Mg could occur with relatively more 35-M SN progenitors than in the disk. The enhanced Ti is not explained by any SN nucleosynthesis predictions. The Ti enhancements seen in bulge stars present a nice qualitative explanation for the well-known phenomenon that the spectral type of bulge M giants is later than disk M giants with the same temperature. Frogel & Whitford (1987) suggested that the later spectral types were due to overall super-metallicity of the bulge stars; McWilliam & Rich (1994) argued that the Ti enhancements

22 524 MCWILLIAM are sufficient to create the stronger bulge M giant TiO bands, without affecting overall metallicity. The enhanced Mg abundances may also explain Rich s (1988) high [Fe/H] results, which were based on measurements of the Mg b lines and assumed that the bulge giants have the solar [Mg/Fe] ratio. Unfortunately, the unusual mixture of α-element abundances for the bulge makes it difficult to use these elements to estimate the bulge formation time scale; the simple picture of SN Ia and SN II implies a different time scale depending on which elements are considered. However, the observed Mg overabundances agree with the predictions of Matteucci & Brocato (1990) and a rapid formation time scale for the bulge. Terndrup et al (1995) and Sadler et al (1996) analyzed low-resolution spectra of 400 bulge giants and found the average [Fe/H] 0.11 dex, consistent with the result of McWilliam & Rich (1994). The [Mg/Fe] ratios +0.3 dex and dex respectively. Multi-population synthesis analysis of low-resolution integrated light spectra of the Galactic bulge by Idiart et al (1996a) indicated a mean bulge abundance ratio of [Mg/Fe] =+0.45 dex. Using the same technique for elliptical galaxies and bulges of external spirals, Idiart et al (1996b) showed a general Mg enhancement of +0.5 dex. Worthey et al (1992), using single-population models, analyzed spectra of giant elliptical galaxies and found Mg enhancements relative to Fe between +0.2 to +0.3 dex. These results provide supporting evidence in favor of enhanced Mg in the bulge, as claimed by McWilliam & Rich (1994). An obvious question arising from the population synthesis results is whether Ti is enhanced in external bulges and elliptical galaxies. The abundance results for α elements in the bulge show that chemical abundance ratios are a function of environmental parameters. In this regard, further study of the detailed chemical composition of Galactic components will lead to an understanding of how environment affects chemical evolution, which can be used to interpret low-resolution low-s/n spectra of distant galaxies. In particular, it is necessary to check the McWilliam & Rich (1994) results for O, Ca, and Si because results for these three elements are less reliable than for Mg and Ti. Halo Alpha Elements The enhancement of α elements in the halo has been confirmed by numerous studies, both in the field and the globular cluster system (e.g. Clegg et al 1981, Barbuy et al 1985, Luck & Bond 1985, François 1987, 1988, Gratton & Sneden 1988, 1991, Zhao & Magain 1990, Nissen et al 1994, Fuhrmann et al 1995, McWilliam et al 1995a,b). Because α-element yields are predicted to increase with increasing SN II progenitor mass (e.g. Woosley & Weaver 1995), the [α/fe] ratio is sensitive to